Back to AI Lab
ArXiv Paper

Over-Searching in Search-Augmented Large Language Models

Roy Xie, Deepak Gopinath, David Qiu +4January 12, 2026

Summary

This work shows that search‑augmented models often call tools even when search hurts answers or wastes tokens. It introduces a cost‑aware metric and mitigation tricks, so teams can dial back needless retrieval instead of just adding more context.

Related Content